Reply 20 of 25, by SSTV2
- Rank
- Oldbie
It was low res, even pentiums stuggle with hi-res VESA modes 😀 I didn't run it for long, fiddled with different freq. oscillators, benchmarked it a bit and reassembled original setup. I believe 3.3V CPUs could work at 5V for months if used for typical home tasks, but if it ran in a 24/7 server machine, such CPUs would fry in a matter of days or weeks due to accelerated electromigration.
Indeed, most of the time it's better to stick to 40MHz FSB with stock cache, especially in PCI MBs.
I could only give bench results from tests done on UM486V with AMD DX4:
Vid. card - CL-GD5428 (Genoa 8500VL);
Chipset - UMC UM82C482/UM82C481/UM82C206F;
L2 cache - 256kB 20ns, async.
Doom 2 v1.666 time demo 2 (2001 gametics), hud on, high quality, no sound, clean DOS boot:
108.6MHz (54.3x2) - 1785 realtics (L2 read @ 3-2-2-2WS, write @ 1WS / DRAM read @ 1WS, write @ 0WS);
108.6MHz (54.3x2) - 1866 realtics (L2 read @ 3-2-2-2WS, write @ 2WS / DRAM read @ 1WS, write @ 1WS);
120MHz (40x3) - 1930 realtics (L2 read @ 3-1-1-1WS, write @ 1WS / DRAM read @ 1WS, write @ 0WS).
Note, that L1 was in "Write Through" mode all the time. I've just found out, that there is an internal pull-down on L1 cache selection pin of AM486 DX4-100...
I've used this pinout diagram from AMD's datasheet of DX2/4 CPUs:
pin R17 "CLKMUL" sets the multiplier, and pin B13 (which I've missed) sets either WB (high) or WT (low) L1 cache mode. Any idea how much of a performance boost would it be in WB mode @ 120MHz? Would it increase FPS at least by 2 in Doom's timedemo?
Edit: corrected benchmark results.